prints a string with an appropriate header

convert dictionary to object with keys as attributes should implement with a class inheriting dict but would cause leak in 2.7 ignores non dict ob

bits per dimension metric

container for tensorflow config

forward pass, must implement

implement for shared method to collect variables

samples all latents required to generate data(s)

converts a continuous log density to a lower bound on the log probability in respect to uniform dequant, assumes values are compressed within [0,1]

injects a mask into a tensor for 2d convs to indicate padding locations

generates kernel for haar wavelet downsampling kernel should be inverse of itself

latent variable is diagonal unit gaussian

inverse pass, must implement

detects/manages files created during previous runs assumes log file is always created if checkpoints exist future: add option to always train fres

computes shapes of latents given data shape assumes we split the latent in half every squeeze except last

implement forward pass

get variables for training, must implement

inverse pass

inverse pass

implement simple way to run layers

default to call forward

auto scoping for TF 1.0 compatibility collection mechanism for simple layer tracking